Research Proposal

What is ESG and Why It Matters

• E: Environmental — climate risk, emissions, energy use
  
• S: Social — labor rights, diversity, community impact
  
• G: Governance — board structure, executive compensation, ethics
  
• ESG evaluates non-financial risk and ethical standards
  
• Relevance: ESG data is unstructured and strategic → ideal for ML and causal inference

Overview

Three empirical papers on ESG and financial markets:

1. ESG Disclosure and Price Informativeness
   → Do disclosures improve market efficiency?

2. ESG in Asset Pricing with Deep Learning
   → Do ESG features improve stock return forecasts?

3. Institutional ESG Consistency
   → Do institutions align ESG rhetoric with their actions?

All papers combine machine learning and causal inference with financial and text data.

Paper 1 – Goals and Motivation

Main Goal:
To examine whether ESG disclosures improve the efficiency of financial markets.

Key Questions:
- Do ESG disclosures lead to lower return volatility?
- Do they reduce bid-ask spreads?
- Do they accelerate price discovery?

Why It Matters:
- ESG reporting is increasing but varies in quality.
- Prior studies focus on the presence of disclosure, not the content.
- Investors and regulators need to know what kind of ESG reporting actually informs markets.

Paper 1 – What I Will Do

• Collect ESG disclosure text from 10-Ks and sustainability reports
  
• Use BERT/RoBERTa to extract:
  • Sentiment, specificity, readability, topical focus
    
• Apply contrastive learning to detect boilerplate vs meaningful content
  
• Estimate market effects: volatility, spreads, price delay
  
• Identify causal effects using IV (e.g., regulatory changes)
  
• Combine ML with inference using prediction-powered methods

Paper 1 – Expected Results

Expected Results:
- Forward-looking, specific ESG disclosures will reduce:

• Return volatility
  
• Bid-ask spreads
  
• Price delay

Why?
- Christensen, Serafeim, & Sikochi (2022):
High-quality ESG disclosures improve liquidity and transparency.
→ Content quality, not just disclosure presence, drives outcomes.

Paper 1 – My Contribution

• Apply transformer NLP (BERT) to ESG reporting content
  
• Use contrastive learning to evaluate substance vs boilerplate
  
• Estimate causal effects using IV and prediction-powered inference [Zrnic and Candès (2023)]
• Advance ESG market microstructure research using text-based methods

Paper 2 – Goals and Motivation

Main Goal:
To test whether ESG features improve stock return prediction in a machine learning framework.

Key Questions:
- Can ESG data improve asset pricing models?
- Are ESG-informed portfolios more profitable and stable?

Why It Matters:
- Traditional models (e.g., Fama-French) assume linearity.
- ESG data is high-dimensional, likely nonlinear, and underused.
- ML has improved pricing models (Gu et al., 2020) but rarely includes ESG.

Paper 2 – What I Will Do

• Build firm-month panel: returns, fundamentals, ESG (CRSP, Compustat, Bloomberg, LSEG)
  
• Apply ML models: Lasso, XGBoost, Neural Networks
  
• Add ESG features: emissions, board diversity, controversies
  
• Evaluate model performance: out-of-sample R², Sharpe ratio, turnover
  
• Interpret models using SHAP values and PDPs

Paper 2 – Expected Results

Expected Results:
- ESG features will enhance predictive power
- ESG-augmented models will yield:
- Higher Sharpe ratios
- Better R²
- Lower turnover

Why?
- Gu, Kelly, & Xiu (2020):
ML models outperform linear ones in return prediction.
→ Adding ESG may reveal alpha related to risk or preferences.

Paper 2 – My Contribution

• First study integrating ESG into full ML asset pricing pipeline
  
• Demonstrates ESG alpha using interpretable ML
  
• Bridges predictive power and economic interpretation
  
• Offers practical tools for investors managing ESG portfolios

Paper 3 – Goals and Motivation

Main Goal:
To assess whether institutional investors align ESG claims with their real behavior.

Key Questions:
- Do ESG statements match actual voting and portfolio decisions?
- Do inconsistencies affect investor flows?

Why It Matters:
- ESG credibility is under fire — greenwashing is a top concern
- No scalable method exists to measure ESG consistency
- Investor trust depends on aligning talk and action

Paper 3 – What I Will Do

• Collect ESG policy statements, proxy votes (N-PX), and holdings (13F)
  
• Apply transformer NLP to extract ESG pledges by theme
  
• Link actions to claims using LinkTransformer + HDBSCAN
  
• Build ESG Consistency Index
  
• Use DiD regressions to test effect on fund flows (CRSP)

Paper 3 – Expected Results

Expected Results:
- Many funds will show inconsistencies across statements, votes, and holdings
- Inconsistent funds will lose capital from ESG-conscious investors

Why?
- Raghunandan & Rajgopal (2022):
ESG-branded funds often vote against ESG proposals.
→ This disconnect signals symbolic rather than substantive commitment.

Paper 3 – My Contribution

• Create the first scalable ESG Consistency Index
  
• Use NLP + record linkage + clustering to evaluate ESG behavior
  
• Quantify the financial consequences of ESG inconsistency
  
• Help investors and regulators hold ESG funds accountable